Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76046.63191128224,
-      "required_pages_per_second_upper_estimate": 0.006591529315643466
+      "required_token_positions_per_second": 76012.64047276191,
+      "required_pages_per_second_upper_estimate": 0.006588583023903037
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9505.82898891028,
-      "required_pages_per_second_upper_estimate": 0.0037920173084850326
+      "required_token_positions_per_second": 9501.580059095239,
+      "required_pages_per_second_upper_estimate": 0.003790322346854651
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19011.65797782056,
-      "required_pages_per_second_upper_estimate": 0.004398012479416172
+      "required_token_positions_per_second": 19003.160118190477,
+      "required_pages_per_second_upper_estimate": 0.004396046649147975
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28517.48696673084,
-      "required_pages_per_second_upper_estimate": 0.0017365653365830128
+      "required_token_positions_per_second": 28504.740177285716,
+      "required_pages_per_second_upper_estimate": 0.0017357891239830404
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28517.48696673084,
-      "required_pages_per_second_upper_estimate": 0.002195355088155342
+      "required_token_positions_per_second": 28504.740177285716,
+      "required_pages_per_second_upper_estimate": 0.002194373805018485
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28517.48696673084,
-      "required_pages_per_second_upper_estimate": 0.3540917462375162
+      "required_token_positions_per_second": 28504.740177285716,
+      "required_pages_per_second_upper_estimate": 0.3539334737733677
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07752646051569911,
-    "producer_p95_seconds": 0.21125603599830356,
-    "gpu_update_seconds": 0.9695103931231667,
-    "producer_batches_per_second": 12.898821813198861,
-    "gpu_batches_per_second": 1.0314484579980774,
-    "producer_headroom": 12.505541806940107,
+    "producer_batch_seconds": 0.08209260006246666,
+    "producer_p95_seconds": 0.20551990899912198,
+    "gpu_update_seconds": 0.9699439401321603,
+    "producer_batches_per_second": 12.181365911654288,
+    "gpu_batches_per_second": 1.0309874196066882,
+    "producer_headroom": 11.815242048541545,
     "sustainable": true,
-    "buffer_seconds": 60.109644373636336,
+    "buffer_seconds": 60.13652428819394,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376788.25974089396,
-    "upload_bytes_per_second": 6017077.810174047,
+    "generation_bytes_per_second": 420340.3110506688,
+    "upload_bytes_per_second": 5971646.863384404,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.84998935861697,
+    "estimated_full_shards_per_hour": 90.19524572982834,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -538,6 +538,18 @@
       "bytes": 873551,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 0.8877478260001226,
+      "bytes": 6165145,
+      "pages": 35,
+      "shards": 2
+    },
+    {
+      "seconds": 0.9784168599999248,
+      "bytes": 2018917,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -98,3 +98,4 @@
 {"step": 49800, "loss": 0.5993867069482803, "lr": 0.0003, "grad_norm": 0.13161946458735313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7288087450, "tokens_per_second": 150009.94783579005, "data_wait_s": 9.265000699087977e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 49900, "loss": 0.6517329514026642, "lr": 0.0003, "grad_norm": 0.13602953573908014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7302896789, "tokens_per_second": 151602.71017741144, "data_wait_s": 1.0196001312579028e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 50000, "loss": 0.678381010890007, "lr": 0.0003, "grad_norm": 0.14200276719499727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7317670920, "tokens_per_second": 151211.93643269406, "data_wait_s": 8.727000022190623e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 50100, "loss": 0.661159873008728, "lr": 0.0003, "grad_norm": 0.16064657262487617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7332387355, "tokens_per_second": 139554.97944974128, "data_wait_s": 1.0337000276194885e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2871b8a9a7c866ad0a1776de66f3a102539f62fca599ed733890d01c103d3629
-size 213531
+oid sha256:27fade22a59f62f375dec37e0d870b7b721001f978f244199065343a31a5618f
+size 215649